In my opinion there's a wide open array of approaches from control that can help with this. Learning for Control is a new conference that looks at this very topic.